9600 np volt mod

With all my exams over, I finally found the time to do what I always wanted to: volt-mod my 9600 np. I first changed the stock heatsink (passive cooling) with a trimmed down Titan D5TB (aluminium heatsink) with a 60mm fan mounted over it. I had to improvise a holding mechanism since the heatsink was too big. I used screws with small springs attached to them to absorb any shocks. I used Arctic Alumina on the GPU. Then I soldered a 20k pot to the 5 and 7 pins of the IC called ISL6522CB (found on all 9600, pro and non-pros). Varying its resistance, I was able to modify the stock vgpu from 1.19V to max of 1.40, after which artifacts started to appear. At 1.34V I was able to overclock the GPU up to 510 w/o artifacts or crashing, but no further (artifacts would appear if I increased the voltage, or it would simply crash if I overclocked more). With the maximum overclock on the memory being at about 230 MHz (w/o any modding, *yet* ) and the GPU running 500 MHz, I got 11500 pts in 3DMark2001SE and 3150 in 3dmark03. I am currently looking for a mod for the vmem, since I already have heatsinks (made from yet another Titan D5TB) but I haven't attached them yet because I'm looking for a safe and almost permanent solution to do it. I'm expecting at least 250 MHz out of my memmory because they're 4ns Hynix chips.

To see pictures of the mod click here.

========================================
Up till here is a copy-paste from another forum where I usually post, but vmods are not very frequent over there.


Now, with my vgpu done I'd like to mod the vmem.

I checked all the chips on my card and I noted down all the info on them, as can be seen in the following 2 pictures. I tried looking up their datasheets, and was able to find what I wanted for only 2 of them - ISL6522CB and CBT3257 - first of which was used to mod the vgpu; the 2nd I have no clue what it is used for. Now what I'm asking is either tell me what I need to do, or just tell me where I can find info on the rest of these chips. I suspect (though I am clueless about it) that the two IOR chips are responsible for the vmem.

Thanks for reading all this.

back of the card:

front of the card:

For the full size pictures click here: front and back.

Nice mod doing it yourself and all....

There are 3 other control chips on the board: DDR bus termination regulator (usually linear), a VDDQ Switching regulator and a VDDR Switching regulator. I will get you datasheets for each of the chips...

25P05AV = Serial BIOS chip
AD66 = nada
CBT3257 - demultiplexer?
ISL6522CB
No match for IOR7413, but IR7413 = MOSFET
R7201 = Another MOSFET

Actually I think you may have RAM that is unmoddable. There are no other voltage regulators so it could be that the RAM is 3.3V - driven directly from the 3.3V line. I had this on my Geforce 4 MX440 - no controller

Ok, I found the other 2 small chips on the backside of the board: IRF7413 and IRF7201.

Now, can you tell me what pins I need to soldier on what chip?

[Edit]

Actually, not true. I have 2 different voltages under 2 capacitators: one gives me 3.35V, the other 3.33V, so I think one is vddr and the other is vddq (I have no clue what I'm saying, I just read that vddq is just a bit lower than the other one )

Plus, I've seen these 8 pin chips being modded on other cards, so I assume they control at least *something*, right?

The bit circled could be a linear DDR regulator, although I haven't seen one that uses MOSFETs. Can you find out any codes on the chips with yellow dots on?

Need sleep mmmmhmhmh...

Ok, looking now. Be back in 10 minutes

Okay, those small 6 pin chips are all marked 431L. The two MOSFETs (I have no clue what a MOSFET is, but I know those are MOSFETs ) are marked APM3055L GF277. I also drew the points where I read 3.35V and 3.30V (varying the AGP voltage from 1.5V to 1.7V did not change the readings at all). Check the image.

You might also want to take a look at this 9000pro vmod. Check out the part about vmem. Sounds like something you said.


This 9100 vmod uses the same 431L (APL431) chip I found on my card. Check it out, maybe it helps you understand how my card was designed.

[Edit]
I took the time and noted down all marked resistors(?), as can be seen in this image. The 431L (APL431?) is in fact a 3 pin chip, but the PCB below has 6 solder points(?), three of which are unused.

There is another 5 pin chip (2+3) called AC25, marked with double rectangles in the image.

Hope all this information can help you determine what needs to be done.

PS: thanks for all your time.

Your memory uses 2 linear voltage regulators each connected to a MOSFET for it's voltages. The 3 little chips have a maximum current of 150mA so they definitely need a MOSFET or something there. What is really odd is that the application circuit uses an NPN transistor instead of an N-type MOSFET. The thing is, it is feedback driven, so that makes it moddable .

In theory putting another resistor from the Vref pin to ground will increase the voltage, but first you need to know what voltage you are increasing.

What I really need to know is: What are the codes on the RAM chips? The datasheet there should give us clues as to what the supply voltages are (I still suspect that they will be 3.3V as the RAM is quite slow - 200MHz.)

If you Vmod the ram you are definitely going to need cooling on it.

Cooling is not a problem, but I do want to do the vmod first, so I can clean the whole thing up and tidy it a bit before I apply the ramsinks (which are huge btw ).

My ram is Hynix HY5DV281622DT-4. It is rated for 250 MHz (4ns) and what intrigues me is that I can only reach 230-235 MHz w/o artifacts, so I suppose they're undervolted on purpose.

So, should I assume that the two 413L are vdd and vddq? should I attempt to mod them? I just bought a 50k trimpot (really nice - the only one I could find though) and I'm pretty eager to use it

What do you think?

My guess is that they are VDD and VDDQ. Finding VDD is important since VDDQ needs to be lower than VDD by 0.3V. put it higher than and strange stuff will happen. Increasing VDD shouldn't cause any problems though.

Standard operation voltages are 3.3V for VDD and 2.5V for VDDQ. I assume that the VDD voltage is driven from either 5V or 12V. VDDQ on my Radeon 9600XT is driven from 12V while VDD is 5V driven.

To measure these voltages I think a good place to look would be on the MOSFETS. Can you measure the voltages at the points shown with respect to ground?

The measurements are:

What now? Do I just shunt the VREF on the 2 small chips? Do I have to use a specific resistor value with it? I'm thinking of starting with 50k ohms.

Leave the lower MOSFET alone - it's VDDQ and probably doesn't affect the overclock that much.

Ok this bothers me. Looking at the datasheet from several companies (but they look like clones of the same to me) the internal circuit of the regulation chip looks like it has the op-amp wired up the wrong way round in the diagram. Putting the feedback into the non-inverting input will cause the system to oscillate as it is positive feedback. I will investigate further.

EDIT: Ok you can get rid of the oscillations in a transistor circuit with a capacitor between A and REF, but I still don't see how this circuit works with a MOSFET

Have you found anything new yet?

What do you suggest I do now? Would there be any risk if I shunted the Vref pin?

If those voltages are very near to what is added on the diagram, then that is the chip to mod. If you have the dexterity and a really small soldering tip then solder a wire to the VREF pin on the chip - find which one it is from the datasheet you posted. Otherwise you may have to use pomona grabbers as it is really small.

The datasheet is in agreement with this: Vo = 2.5x(1+R1/R2)

reducing R2, the one between Vref and ground increases the voltage.